Genesis Troubleshooting

Kubernetes services failures

Before the Armada manifests are applied, the script will bring basic kubernetes services online by starting docker containers for these services.

One of the first services to be brought up is the kubernetes API. If it fails to come up, you may see a repeated error as follows from the script:

.The connection to the server apiserver.kubernetes.promenade:6443 was
refused - did you specify the right host or port?

Check that the hostname in your Genesis.yaml matches the hostname of the machine you are trying to install onto. If they do not match, change one to match the other. If you change Genesis.yaml, then re-generate the Promenade payloads.

If the hostnames match, check the container logs under /var/log/pods to see the reason for the provisioning failure. (kubectl logs function will not be available if the API container is not running).

Armada failures

When executing, you may encounter failures from Armada in the provisioning of other containers. For example:

CRITICAL armada [-] Unhandled error: armada.exceptions.tiller_exceptions.ReleaseException: Failed to Install release: barbican

Use kubectl logs on the failed pod to determine the reason for the failure. E.g.:

sudo kubectl logs barbican-api-5b8bccdf8f-x7sld --namespace=ucp

Other errors may point to configuration errors. For example:

CRITICAL armada [-] Unhandled error: armada.exceptions.source_exceptions.GitLocationException: master is not a valid git repository.

In this case, the git branch name was inadvertently substituted for the git URL in one of the chart definitions in bootstrap-armada.yaml.

Post-run failures

At its conclusion, the genesis script will output the list of containers provisioned and their status, as reported by kubernetes. It is possible that some containers may not be in a Running state. E.g.:

ucp   promenade-api-6696769cd-qwpzf   0/1   ImagePullBackOff   0   10h

For general failures, kubectl logs may be used as in the previous section. In this case, it was necessary to run kubectl describe on the pod to get the details of the image pull failure. E.g.:

kubectl describe pod promenade-api-7dc54d47c-qw27m --namespace=ucp

In this particular incident report, the problem was a missing certificate on the bare metal node which caused the image download to fail. Installing the certificate, restarting the docker service, and then waiting for the container to retry resolved this particular issue.